A Review of Generalization in Additive Manufacturing Optimization

Additive manufacturing (AM) has transitioned from prototyping to functionalpart production across aerospace, biomedical, automotive, energy,tooling, and consumer sectors. However, achieving consistent part qualityremains challenging due to the strong sensitivity of AM processes to a multitudeof interdependent parameters. Over the past decade, thousands ofstudies have optimized process parameters for specific materials, machines,and geometries. Yet, these optimized solutions overwhelmingly lack generalizability:parameters that yield good results on one machine, geometry, AMtechnology, or material often fail when transferred to another context.This review provides the first unified, cross-technology examination ofthe generalizability problem in AM process-parameter optimization. Wesynthesize evidence from polymer, metal, and ceramic AM processes— includingFused Filament Fabrication (FFF), Stereolithography (SLA), SelectiveLaser Sintering (SLS), Laser Powder Bed Fusion (LPBF/SLM), ElectronBeam Melting (EBM), Directed Energy Deposition (DED), and binderjetting. The review identifies root causes of non-generalizability, includingmachine-to-machine variation, material thermal and rheological inconsistency,geometry-scaling effects, environmental sensitivity, multiscale multiphysicsinteractions, and the lack of standardized benchmark datasets.We critically evaluate optimization approaches ranging from design-ofexperiments(DOE) and regression models to machine learning (ML), deeplearning, Bayesian optimization, simulation-driven optimization, and multifidelitymodelling. The evidence reveals that most models fit locally but failto transfer across new materials, geometries, and machines due to limiteddatasets, local search spaces, and insufficient physics incorporation. Emergingdirections such as meta-learning, transfer learning, physics informed machinelearning, multi-fidelity data fusion, closed-loop control, and digitaltwins are proposed as pathways toward generalized, scalable optimization.The review concludes by outlining research gaps that must be overcome forAM to achieve fully autonomous, machine-independent, quality-assured manufacturing.